High-Dimensional MR Reconstruction Integrating Subspace and Adaptive Generative Models.

Area of Science:

• Medical Imaging
• Computational Imaging
• Machine Learning in Imaging

Background:

• High-dimensional Magnetic Resonance (MR) imaging presents reconstruction challenges.
• Existing subspace-based methods have limitations in capturing complex image variations.

Purpose of the Study:

• To develop an integrated method for high-dimensional MR image reconstruction.
• To leverage generative models as adaptive spatial priors within subspace frameworks.

Main Methods:

• Proposed a formulation synergizing subspace models, adaptive generative image priors, and sparsity regularization.
• Developed a pretraining and subject-specific network adaptation strategy for generative representations.
• Introduced an iterative algorithm for joint updates of subspace coefficients and generative model latent spaces.

Main Results:

• Demonstrated improved performance in accelerated MR parameter mapping.
• Showcased enhanced results in high-resolution MR spectroscopic imaging.
• Outperformed state-of-the-art subspace-based methods in both applications.

Conclusions:

• The method offers a novel approach to high-dimensional MR image reconstruction.
• Integrating adaptive generative models provides effective data-driven spatial constraints.
• Highlights the potential of combining data-driven priors with low-dimensional modeling for advanced imaging.